About the survey and study population
The DHS survey program (www.measuredhs.com) has been operating in Bangladesh since 1993. The present study was based on data collected from the sixth wave of Bangladesh Demographic and Health Survey (BDHS) carried out in 2011. The survey is cross-sectional in nature and the sample population is nationally representative. The survey was conducted under the authority of the National Institute of Population Research and Training (NIPORT) and was implemented by Mitra and Associates, a well-known research institution in the country. Technical and financial assistance were provided by ICF International of Calverton (Maryland, USA) and United States Agency for International Development (USAID) respectively [13].
Bangladesh is the eighth most populous country in the world and third most populous in South Asia. It is also the most densely populated country in the world (excluding the city-states) with about 1015 inhabitants/km2 [13]. Due to rapid population growth, family planning policies to curb fertility rates have been in place throughout the country since its independence in 1971. The country is divided into seven administrative divisions (Barisal, Chittagong, Dhaka, Khulna, Rajshahi, Rangpur, and Sylhet) and survey was carried out in all these divisions encompassing both rural and urban population. The details of survey design, sampling method, data collection and distribution have already been described elsewhere [13, 19]. Briefly, the survey employed a two-stage stratified sampling of the households in a systematic way to ensure that the sample is nationally representative. In total 18,222 ever married women ageing between 12 and 49 years were selected for interview, of whom 17,842 were finally surveyed (response rate of 98%). The principle objectives of the survey were to provide most recent scenario of demographic (e.g. fertility, infant & maternal mortality rates), socioeconomic (e.g. literacy, employment, wealth status, food security), health indicators (e.g. malnutrition, rates of contraceptive use, skilled birth assistance, health literacy).
Variable selection
The primary outcome variable of this study was pregnancy intention status among women for their last pregnancy. Though it is recognised among researchers that the concept of unintended pregnancy is a complex and nuanced one [17], it generally includes pregnancies described as mistimed or unwanted, and is usually used as a binary outcome (intended/unintended or wanted/unwanted). In this study it was measured by the response to the question on pregnancy intendedness and was dichotomized followingly: intended = wanted then, and unintended = wanted later/wanted no more.
Unmet need for contraception was the main exposure variable in this study. It was measured based on contraception utilisation status among participants who did/did not report desire for spacing or limiting childbirth. Unmet need for contraception refers to non-utilization of contraception measures among women who are fecund and sexually active, not want any more pregnancies or want to delay the next pregnancy. The concept of unmet need points to the gap between women’s reproductive intentions and their contraceptive behaviour. Participants who wanted to space/limit childbirth but not using contraception were considered as having unmet need for contraception.
For the selection of relevant covariates, we conducted a literature search in prominent medical databases for studies on unintended pregnancy. Based on literature review, and availability in the BDHS dataset, the following variables were selected for this study: age (<25/≥25); type of residence (urban/rural); educational attainment of respondent: nil (no formal education), primary (1–5 years of formal education); secondary/higher (>5 years of formal education); educational attainment of husband nil (no formal education); primary (1–5 years of formal education); secondary/higher (>5 years of formal education). husband’s occupation: blue collar (includes employments in farming, construction; white collar (includes employments in service, teaching, health sector, business); employment (yes/no); sex of household head (male/female); age at first birth (<18/≥18); ideal number of children (1/>1); has a say in own healthcare decision (yes/no); ever had a terminated pregnancy(yes/no); currently using any contraception (yes/no); unmet need for contraception(yes/no); decision maker for using contraception (yes/no); household wealth status (poor/middle/rich). DHS studies employs principal component analysis method to measure wealth index of households based on ownership of household assets e.g. durable goods (television, bicycle), household characteristics (sanitation facilities, construction materials) [13]. The factor scores are summed and standardized for each household which places them in a continuous scale based on relative wealth scores. Then the scores are categorized into quintiles where each households fall into a category, with the lowest scores representing the poorest and highest representing the richest households [19].
Data analysis
The BDHS dataset contain information on wide range of variables. The dataset was checked to include only those participants for whom all information necessary for the present study were available. Prior to analysis, data were weighted by sample weights to generate population estimates. Descriptive analysis (frequency distribution) was performed to show the basic characteristics (e.g. demographic, socioeconomic, contraceptive use) of the sample population. Pearson’s χ
2 tests as well as cross tabulation were performed to show the group differences in terms of pregnancy intentions across the explanatory variables. At this stage, the explanatory variables were checked for multicollinearity to ensure validity for further analysis. Variables which showed statistically significant association (p < 0.05) in the χ
2 tests were included in the multivariable regression analysis. Given the complex clustered nature of the survey, we employed generalised estimating equation (GEE) method which serves a reliable tool for dealing with clustered data [19, 20]. GEE is a commonly employed statistical approach to fit a marginal model for clustered data analysis in clinical trials and biomedical studies [20, 21]. Adjusted odds ratios (OR) with 95% confidence intervals (CI) were calculated to measure the strength of associations between pregnancy intention status and the response variables in the model. A two-tailed p-value of <0.05 was used to assess the statistical significance for all tests. All analyses were carried out with the Mac Version of IBM SPSS Statistics 21.